If AI Must Pay Taxes, How Much Should It Owe?

Opening

Hello, subscriber. This is Oswarld’s Knowledge Talking.

Two weeks ago, Block, the fintech company led by Jack Dorsey, laid off 40% of its workforce — about 4,000 people. Dorsey explained it this way: “AI tools have changed the very way we build and run the company.” What’s striking is how the market reacted. The stock jumped 24%. That’s roughly $6 billion added to market capitalization in a single day — about $1.5 million in added corporate value for every employee let go.

The same week, Andrew Yang — former U.S. presidential candidate and entrepreneur — said in a CNBC interview: “We should stop taxing labor and start taxing AI agents.” This isn’t just political rhetoric. As Block’s case shows, when AI displaces human jobs, the very foundation of government tax revenue starts to wobble. Today I want to unpack this structural problem. (Personally, Yang is one of the Korean-Taiwanese-American politicians I follow with particular interest.)

The Quiet Revolution: Downsizing Without Layoffs

AI’s shock to the labor market is showing up first not as mass layoffs, but as a disappearance of new hiring. Dario Amodei, CEO of Anthropic, warned in a May 2025 Axios interview that “up to 50% of entry-level white-collar jobs could be automated within the next one to five years.” He also projected that unemployment could spike to 10-20%. What makes this warning notable is that it came from the CEO of the very company building the automation technology in question.

Let’s look at the numbers. According to Federal Reserve Bank of New York data, the U.S. unemployment rate for recent college graduates (ages 22-27) stood at 5.7% in Q4 2025 — above the overall unemployment rate of roughly 4%. For the first time in history, the unemployment rate for college graduates has exceeded the overall rate. The underemployment rate¹ hit 42.5%, the highest level since 2020.

Goldman Sachs Research dug into the cause: in the industries where college graduates typically find work — information technology, finance, and professional services — average monthly job growth between 2023 and 2025 came in at -9,000. Over the same period, industries dominated by non-college-graduate workers averaged +12,000 jobs per month. In other words, new hiring is freezing up first in precisely the industries where AI adoption is most aggressive.

Andrew Yang put it bluntly: “The easiest person to fire is the person you haven’t hired yet.”

The Tax Paradox: When Labor Disappears, So Does Tax Revenue

This is where the tax structure problem comes in. More than half of U.S. federal tax revenue comes from individual income tax — about $2.6 trillion in 2025, roughly ₩3,500 trillion in Korean won. Add payroll tax, and labor-related taxes account for about 75% of total federal revenue.

So what happens when AI replaces labor within this structure? Corporate profits rise, but as payrolls shrink, both the income tax base and the payroll tax base weaken simultaneously. Think back to Block: 4,000 people lost their jobs, and the stock rose 24%. It’s a textbook pattern of productivity gains flowing to capital owners.

Andrew Yang takes direct aim at this structure. “We tend to tax the things we want less of. But right now we’re taxing labor — in an era when we need more of it, not less.” His proposal is clear: stop taxing labor, and tax AI agents instead.

Interestingly, Dario Amodei has staked out a similar position. He’s proposed a “token tax” that would channel 3% of AI companies’ revenue into government redistribution programs. When an AI company’s own CEO says “please tax us,” it’s worth pausing — as Yang put it — to ask, “since when has that ever happened?”

Agent Tax: Easy to Say, Hard to Design

So what does it actually mean to “tax AI”? This is where things get complicated.

Current proposals for taxing AI fall broadly into three categories.

First, profit-based taxation — levying additional corporate tax on excess profits generated by AI. This is the direction recommended by the IMF in its 2024 report. The IMF argued that “taxing AI directly would be difficult to administer and could stifle innovation,” advising instead that governments strengthen capital gains taxation and revisit existing tax incentives that encourage labor substitution.

Second, usage-based taxation — taxing AI model usage, similar to Amodei’s token tax. According to a January 2026 analysis from the Brookings Institution, this approach functions much like existing consumption taxes when applied to end consumers, but when applied to business-to-business (B2B) transactions, it can trigger a cascade effect² where taxes compound at each stage — undermining the competitiveness of AI products.

Third, task-substitution-based taxation — a “task tax” levied per instance when AI or robots replace a specific human task. Zak Kidd, founder of the AI company AskHumans, has been actively pitching this model to U.S. state governors. Take a hotel as an example: when a housekeeping worker earning $28 an hour is replaced by a robot costing $2 an hour, this model would tax back a portion of that difference.

All three approaches share one fundamental difficulty: how do you separate AI’s contribution from a human’s? As Kidd points out, there’s still no clear answer to the question of “where you draw the line between where AI stops and human judgment begins.”

So Could Universal Basic Income Be the Answer?

The question of where agent tax revenue should go naturally leads to universal basic income (UBI).

Andrew Yang is the figure who proposed a “Freedom Dividend” of $1,000 a month ($12,000 a year) for every American adult during his 2020 presidential run. At the time, an analysis by the Tax Foundation found that a 10% value-added tax (VAT) alone wouldn’t be enough to fund it, and that the marginal tax rate on labor would need to rise by about 8.6 percentage points — potentially shrinking GDP by 3% over the long run.

But by 2026, the picture has shifted. The productivity surplus generated by AI is emerging as a new potential funding source. Block’s annual cost savings from cutting 4,000 jobs are expected to run into the hundreds of millions of dollars. The core logic of an agent tax is to funnel a portion of that surplus back into society.

Korea isn’t exempt from this discussion either. According to an IMF report published in 2025, roughly half of all jobs in Korea are exposed to AI’s impact — with particularly high exposure among office workers, women, young people, and highly educated workers. Research from the Korea Development Institute (KDI) reaches a similar conclusion, finding that the spread of AI adoption is mainly reducing job opportunities for young workers.

President Lee Jae-myung raised the vision of an “AI basic society” at the APEC summit, and domestically, discussions of a robot tax, AI profit-sharing schemes, and universal basic income are gaining momentum. In the end, the core question is the same everywhere: how do you raise the funds, and how do you distribute them?

Oz’s Lens

Honestly, I think the structural question behind the term “agent tax” matters more than the name itself. In the early stages of adopting any technology, “efficiency” gets all the attention — but where that efficiency’s gains actually flow is a completely separate matter. Block’s case shows this precisely. The layoff of 4,000 people meant a $6 billion increase in value for shareholders, but for the workers let go, it meant 20 weeks of severance pay.

What catches my attention is a point that both the IMF and Brookings make in common: taxing AI directly is technically difficult and, done poorly, could stifle innovation. Instead, the realistic first step is to strengthen capital gains taxation and redesign existing tax incentives that overly encourage labor substitution.

Ultimately, this isn’t a tax problem — it’s a problem of redesigning the social contract. The 20th-century welfare state was built on a virtuous cycle: labor → income → consumption → tax revenue. AI is weakening the very first link in that chain — labor — and simply patching the remaining links isn’t a fundamental solution.

Of everything Andrew Yang said, what I find most meaningful isn’t the tax proposal itself, but this line: “America’s implicit social contract is breaking down.” It’s a diagnosis of a world where college graduates face higher unemployment than non-graduates, and where a company’s stock jumps when it fires employees — a world where the promise between labor and reward is coming apart. This isn’t just an American problem. The same pattern is visible in Korea’s youth employment market too.

Closing

When AI displaces labor, the tax base weakens — and that threatens the sustainability of welfare systems. Whether it’s called an “agent tax,” a “token tax,” or a “task tax,” it’s becoming increasingly clear that we need some mechanism to channel AI’s productivity surplus back into society. But the design is still in its early stages, and the technical question of “what exactly should be taxed” matters just as much as the question of “will AI eliminate jobs.”

The framework Brookings offered in its 2026 analysis is useful here: distinguish between reforms needed right now and reforms needed for a future where AI operates with far greater autonomy. For now, the priority is revisiting tax incentives that encourage labor substitution and strengthening capital gains taxation. Taxing AI directly is a conversation for later.

References & Further Reading

• Andrew Yang, CNBC Squawk Box interview, March 12, 2026. — The interview that sparked today’s newsletter. You can hear the context behind the agent tax proposal directly.
• Dario Amodei, Axios interview, “AI jobs: Sleepwalking into a white-collar bloodbath”, May 28, 2025. — The original source for Amodei’s 50% automation warning and token tax proposal.
• IMF, “Broadening the Gains from Generative AI: The Role of Fiscal Policies”, Staff Discussion Note SDN2024/002, June 2024. — The IMF’s comprehensive analysis of fiscal policy in the AI age.
• Anton Korinek & Daniel Martin Katz, “The future of tax policy: A public finance framework for the age of AI”, Brookings, January 8, 2026. — The most systematic analysis available on the technical design of an agent tax.
• Federal Reserve Bank of New York, “The Labor Market for Recent College Graduates”, Q4 2025 data. — The original data source for college graduate unemployment and underemployment rates.
• Goldman Sachs Research, “Why Is the College Graduate Unemployment Rate Higher Than Usual?”, February 2026. — A report analyzing the industry-specific causes of worsening graduate employment.
• KDI (Korea Development Institute), “Changes in the Labor Market Due to Artificial Intelligence and Policy Directions”, Research Report 2023-03. — A key domestic empirical study of AI’s impact on Korea’s labor market.

The author, Kwangseob Ahn, is a professor of business administration at Sejong University and lead consultant at OBF (Oswarld Boutique Consulting Firm). He teaches statistics and data analysis — business data management and business analytics — while leading GTM and AI strategy consulting in the field, designing the seam between technology and business. He has published academic research on a memory architecture for AI dialogue systems (HEMA) and runs Daily Arxiv, a daily curation of global AI papers. He holds a master’s from Korea University’s Graduate School of Technology Management and a KMBA. He is the author of Homo Brainless: The People Who Outsource Their Thinking.

Footnotes

1. Underemployment: A situation where a college graduate ends up in a job that doesn’t require a college degree — for example, a business major working at a coffee shop. ↩

2. Cascade Effect: A phenomenon where tax is applied repeatedly at every stage of a supply chain, driving up the final price excessively. Value-added tax (VAT) is designed to prevent this, but new forms of taxation like a token tax could reintroduce the problem. ↩